Climatic Atlas of Clouds Over Land and Ocean (website)

 

by Ryan Eastman, Stephen G. Warren, and Carole J. Hahn,

 

July 2014

 

 

Introduction

      This website (www.atmos.washington.edu/CloudMap/) contains revised and updated versions of the maps presented in the 1980’s in our atlases of cloud climatological data obtained from visual observations at the Earth’s surface (Warren et al. 1986, 1988).  The cloud averages presented on these maps come from our digital archive of gridded land and ocean cloud climatological data which, along with the documentation, are now available (Hahn and Warren 2007).

 

      Maps are given for total cloud cover, clear-sky frequency, and the average amounts of nine cloud types within the low, middle, and high levels of the troposphere.  [The “amount” of a cloud type is the fraction of the sky-hemisphere covered by that type.]  Maps of precipitation frequency are also included.  Monthly, seasonal, and annual averages are given for both daytime and nighttime.  Land and ocean data were analyzed separately, and are mapped separately for most quantities.  Digital values for each map can be downloaded from this website.

 

      The climatology presented in these maps supersedes our first cloud climatology in that it is based on 26 years (39 for trends) of data rather than 11 for the land, and 44 years (55 for trends) rather than 30 for the ocean.  Also, more cloud types are distinguished in the low and middle levels, and the analysis procedures have been improved, particularly as regards the nighttime observations.  

 

      Two grid sizes are used to display the cloud averages.  Most data are given at 5-degree latitude-longitude resolution (with coarser longitudinal resolution toward the poles to preserve approximately equal-area grid boxes).  A 10-degree grid (also with coarser longitudinal resolution toward the poles) is used for some ocean data.  

 

      The land maps are based on analyses of 185 million visual cloud observations made at 5388 weather stations on continents and islands over a 26-year period, 1971-1996.  A climatic summary of the cloud variables was previously produced for each station; it is available from the Department of Energy’s Carbon Dioxide Information Analysis Center (DOE-CDIAC) as Numeric Data Package NDP-026D (Hahn and Warren 2003), and the documentation is also available on this website.  To produce the maps in this atlas, the station data were averaged within 5-degree grid boxes.  [In a grid box that contains only one station, the values presented here are the same as those for the corresponding station in NDP-026D.] 

 

      The ocean maps are based on analyses of 50 million cloud observations made from ships over a 44-year period, 1954-1997.  The precipitation maps for both land and ocean are based on 1982-91 data from Hahn et al. (1994).

 

      For each grid box, the digital database includes multi-year annual, seasonal and monthly averages, and the amplitude and phase of annual and diurnal cycles.  Averages are given for total cloud cover, clear-sky frequency, and nine cloud types [five in the low level: fog, stratus, stratocumulus, cumulus, and cumulonimbus; three in the middle level: nimbostratus, altostratus, and altocumulus; and one in the high level: all cirriform clouds combined].  Cloud amounts and frequencies-of-occurrence are given for all types, and average base heights are given for the low cloud types.  We also give the combined amounts of all low-level clouds and of all middle-level clouds.  Separate averages are computed for daytime (06-18 local time) and nighttime (18-06 local time) in addition to averages over the 24-hour period (the “diurnal average”).  Multi-year seasonal averages for each of the eight reporting times per day, and seasonal and monthly averages by year, are given in the gridded archive (NDP-026E) for ocean data and in the land-station archive (NDP-026D) for the land data.  Many of the quantities from the digital archive are mapped on this website, but not all.  In particular, only multi-year averages are mapped here, not the averages for individual years. 

 

      An outline of the quantities mapped is given in Table 1.  The average “amount” for a season is the product of frequency-of-occurrence (f) and amount-when-present (awp).  The meanings of these quantities are illustrated by an example:  If in a particular season altocumulus is reported in 30% of the usable observations and if it covers 40% of the sky when it is present, then f=0.3, awp=0.4, and the seasonal average altocumulus amount is 12%.

 

      For middle and high clouds, we also compute the non-overlapped amount, which is the amount actually seen by an observer from below; i.e., the amount not obscured by lower clouds.  It is analogous to the quantity reported by most satellite-derived climatologies where the amount reported is the amount not obscured by higher clouds.  The sum of non-overlapped amounts is equal to the total cloud cover, whereas the sum of the actual amounts is greater than the total cloud cover because of overlap.  The non-overlapped amount is what is needed to estimate the effects of clouds on the longwave radiation budget at the surface. 

 

 

Table 1.  Quantities computed for the various cloud types.

 

 

Symbol

Amount

Frequency

Amount-when-present

Non-overlapped amount

Base height

Total cloud cover

 

x

 

 

 

 

Fog1

 

x

 

 

 

 

Stratus

St

x

x

x

 

x

Stratocumulus

Sc

x

x

x

 

x

Cumulus

Cu

x

x

x

 

x

Cumulonimbus

Cb

x

x

x

 

x

Nimbostratus

Ns

x

x

x

x

 

Altostratus

As

x

x

x

x

 

Altocumulus

Ac

x

x

x

x

 

All high clouds

Hi

x

x

x

x

 

All middle clouds

 

x

 

 

 

 

All low clouds

 

x

 

 

 

 

Clear sky

 

 

x

 

 

 

Precipitation

 

 

x

 

 

 

 

1By “fog” we mean the frequency of reports of “sky obscured by fog,” so in these reports fog is always overcast, and amount = frequency. 

 

 

      Nighttime averages for cloud variables were obtained using only those reports made under adequate moonlight or twilight, following the illuminance criterion of Hahn et al. (1995).  The random-overlap assumption was used where necessary to estimate the amount-when-present of the upper clouds Ac, As and Hi hidden by lower clouds, and the maximum-overlap assumption was used where necessary for Ns.  [These concepts were discussed by Warren et al. (1986) and by Hahn and Warren (1999).] 

 

 

Data Sources

      The visual cloud observations are reported in the “synoptic code” established by the World Meteorological Organization (WMO 1988).  The data source for this analysis was the "Extended Edited Cloud Report Archive" (EECRA) (Hahn and Warren 1999, updated 2009).  Land-station reports included in the EECRA were originally taken from an archive of the Fleet Numerical Oceanography Center (FNOC) for the years 1971-1976 and from an archive (maintained at NCAR) of the National Centers for Environmental Prediction (NCEP, formerly NMC) for the years 1977-1996.  Because of changes in procedures at NCEP, the NCEP data do not contain cloud-type information after March 1997.  Yearly averages for seasons and months have been updated using the Integrated Surface Database (ISD), but data used for the long-term averages terminates in 1996.  Other problems with these source data sets, that we have dealt with, were discussed by Hahn and Warren (1999).  For the ocean, ship observations in the EECRA were originally obtained from the Comprehensive Ocean-Atmosphere Data Set (COADS; Woodruff et al. 1987).

 

      For the present land climatology we used only those reports that contained cloud-type information (thus the number of reports used for analyses of low cloud types is the same as the number used for total cloud cover, although the number of reports available for middle and high cloud analyses is smaller because low clouds are sometimes overcast).  For the ocean climatology we used reports for total cloud (and clear sky) analysis even if they did not contain cloud-type information, because ship reports are more sparse.  Advantages and problems with these approaches are discussed by Hahn and Warren (2007). 

 

Selection of Stations for the Land Climatology

      Hahn and Warren (1999) listed 11,586 stations for which cloud-type information was reported at least once in the EECRA.   From them, Hahn and Warren (2003) selected 5158 stations that normally report cloud types, had periods of record sufficiently long for analysis of trends, and had sufficient numbers of observations at night for analysis of diurnal cycles.  [On average, nighttime observations contribute only about 30% of the total because many observations are rejected by screening with the illuminance criterion.]  We also rejected a few “renegade” stations that were found to give erroneous observations; most of these are among those listed earlier by Warren et al. (1986).

 

      Of the 1820 5-degree grid boxes, 961 contain some land.  The 5158 selected stations are contained in 715 boxes.  This group of stations left some significant land areas south of 30oN unrepresented (in Africa, South America, Australia and Antarctica).  We therefore added some stations south of 30oN that had periods of record shorter than 15 years, or that made observations only during daytime.  [Stations were added only if they occupied a 5-degree grid-box that was not represented by the original group of 5158 stations.]  The number of stations added was 230, giving a total of 5388 stations selected, and bringing the number of grid-boxes filled to 820.  

 

      A map showing the number of stations in each box is included, as well as a map showing the fraction of land area in each box.  On the land-fraction map, boxes with no land are blank; a zero on this map means a land fraction less than 0.5% but greater than zero. 

 

      It is notable that few stations in the United States offer the full 26 years of synoptic cloud reports.  With the installation of the Automated Surface Observing System (ASOS) in the mid-1990s, most stations in the U.S. stopped reporting cloud observations in the synoptic code around 1995 (Appendix H of Hahn and Warren 1999), despite objections from the climate community (Warren et al. 1991). 

 

Selection of Ship Reports for the Ocean Climatology

      Because ships from the various countries can move all over the globe, ship data present a different set of problems from the land data.  We eliminated certain blocks of data that we found to be problematic.  For example, we rejected observations from the “Historic Sea Surface Temperature” dataset (HSST; years 1952-61) because cloud-type information had been deleted during the construction of that dataset.  In addition, we found that a number of smaller “decks” contained large fractions of reports that were problematic in some way, and we rejected them.  A list of the excluded decks is given by Hahn and Warren (2007). 

 

Averaging Methods

      A brief summary of our averaging methods is given here; detailed descriptions are given in the documentations for the databases (Hahn and Warren 1999, 2003, 2007).

 

      Except for amounts of middle and high clouds, which are discussed below, daytime or nighttime averages for the various cloud variables were computed as the simple average of the available observations.  Values of cloud frequencies and amounts are presented on maps if the number of observations available for computing an average was greater than or equal to a specified minimum (usually 100).  For amount-when-present and cloud-base height, only 30 observations were required, because fewer observations are available for these quantities, and because these quantities are less variable (geographically and seasonally) than cloud frequency.  Most averages presented are multi-year averages for individual months or three-month seasons.  The annual average is computed as the average of four seasons. 

 

      To reduce the "partial-undercast bias" (Warren et al. 1986), frequencies (and amounts-when-present) for upper-level clouds (As, Ac, Hi) were computed only from reports in which the coverage of a lower cloud layer was less than 7/8. 

 

      Because cases in which lower clouds obscure upper levels cannot be used to compute a frequency of occurrence of the upper cloud, our default assumption (Method A) is that the frequency of occurrence of an upper-level cloud type is the same when its level cannot be seen as when it can be seen.  Warren et al. (1988) tested for an upper-level cloud "abstention bias" and concluded that an adjustment might be necessary for Ac and As clouds.  That adjustment (Method B) assumes that the frequency of occurrence of the upper-level cloud type is the same when its level cannot be seen as when it can be seen with the lower cloud amount in the restricted range 3/8 to 6/8.  During the present analysis we looked at this issue again.  For land analyses we found no significant difference between the two methods, so we used Method A.  For the ocean analyses they were different (probably because of the greater frequency of low clouds over the ocean), and we chose Method B.

 

      The synoptic code allows reporting of only two cloud amounts in a single observation (total cloud cover, and amount of the lowest level).  If clouds are present at all three levels, it is thus possible for the amount of a middle or high cloud to be indeterminate even if the cloud is visible.  Therefore we compute awp from the observations for which it can be determined, and we obtain the seasonal average cloud amount as the product of frequency-of-occurrence and awp

 

      A potential bias in the computation of frequencies of cloud types arose due to a change to the instructions for reporting in the synoptic code, beginning in 1982 (WMO 1988), which was intended to reduce the amount of redundancy in a weather report.  This change resulted in a "clear-sky bias" which is discussed in detail elsewhere (Hahn and Warren 1999, 2007).  This bias is eliminated in the present land analysis by using only stations that normally report cloud types.  That restriction is not possible for the ship reports because ship identifiers are not included in the COADS data, so an adjustment factor is applied to cloud-type frequencies over the ocean.  [The adjustment factor is usually close to 1.00 and thus is a minor correction, except in some coastal regions.]

 

      In preliminary versions of the land maps we noticed that several boxes in the region of Indonesia and the western equatorial Pacific, representing island stations, had atypical values of awp computed for middle clouds, and that neighboring boxes often had quite different values of awp.  Observations in this region usually report clouds present at all three levels, thus limiting the computation of awp to a small and unrepresentative sample of reports.  We therefore decided to use, for each middle and high cloud type, the global-mean value of awp for that cloud type when computing cloud amounts for stations in those boxes.  Amount-when-present exhibits far less geographical and seasonal variability than does frequency of occurrence.  Therefore a list of 155 stations (in 47 grid boxes) was made, for which global-mean values of awp are used, together with locally-derived frequencies of occurrence, to compute amounts of middle and high clouds.  This is why the land maps for the Indonesian region show uniform values of awp for Ns, As, Ac, and Hi.  These values are 98% for Ns, 80% for As, 51% for Ac, and 46% for Hi.  This procedure was applied only to the land data in these boxes, because the ocean awp’s did not appear noisy or unrepresentative.

 

Average of day and night (“diurnal average”)

      The diurnal average was computed by one of four methods depending on the variable averaged and the number of observations available. 

(1) For frequencies, amounts, and non-overlapped amounts, it was computed as the average of the day and night values.  This method weights day and night equally; it is the preferred method if sufficient observations are available for both day and night (there are generally fewer usable observations at night because they are screened by the illuminance criterion). 

(2) If the number of observations in either day or night was less than a specified minimum, then an average was computed using all available observations, regardless of time of day. 

(3) For base height, the diurnal average was computed as a weighted average of the day and night values, weighted by the respective frequencies of occurrence.  For example, if cumulus occurs frequently during daytime but rarely at night, then the diurnal average height should be weighted toward the daytime height, as this method ensures. 

(4) For amount-when-present, the diurnal average was computed as diurnal-average amount divided by diurnal-average frequency. 

 

Annual and Diurnal Cycles

      Cloud observations are usually reported every 3 hours or every 6 hours.  For the diurnal cycle, a cosine curve was fitted to the 4 or 8 values for each season; we present maps of the amplitude and phase of the fitted curves.  The “phase” is the local time of maximum (local time at the longitude of the grid-box center).  Similarly, an annual cycle was fitted to the 12 monthly values.  The value of phase is mapped only if there is a significant cycle; to display a value on a map we require that the amplitude of the cycle be at least 1% and the variance accounted for by the cosine curve be at least 30%.  Diurnal cycles are absent from the land maps in some regions (particularly central Africa) where the available stations made reports only in daytime.  Annual and diurnal cycles for ocean data are given on the 10-degree grid because in most parts of the ocean 5-degree boxes do not contain enough observations to resolve these cycles.  Diurnal cycles are discussed by Eastman & Warren (2014)

 

      Annual cycles were computed from the diurnal-average values if all 12 months had at least 75 observations both day and night (96% of the land boxes and 73% of the ocean boxes for total cloud cover).  Otherwise, annual cycles were computed from the daytime values if all 12 months had at least 100 observations for daytime (about 3% of the land boxes and 8% of the ocean boxes).  In a few land boxes (particularly in Antarctica) and many ocean boxes, some months lack sufficient observations, so annual cycles are not computed for those boxes.

 

Zonal and Global Averages

      Zonal averages are listed off the right-hand side of most maps.  The zonal average is computed as the average of all filled boxes in the zone, weighted by their respective land or ocean fractions.  For the land maps this procedure can give biased zonal averages in a few zones that contain significant areas with no stations.  In the desert interiors there are seven boxes with no stations:  two in Australia and seven in North Africa.  Because of the rather homogeneous surface conditions in these regions, we decided to include them in zonal averages, interpolating from neighboring boxes in the same zone to obtain values for the blank boxes.  There are also many missing boxes in Antarctica; the zones 70-75oS and 75-80oS each have only one station in the interior of East Antarctica.  For computing zonal averages these stations are allowed to represent the entire East-Antarctic sector of their zones.

 

      The zonal averages are further averaged (weighted by the land area or ocean area in each zone) to obtain the global average printed at the bottom of each map.  The zonal and global averages shown for awp are averages of awp for boxes, not weighted by f for each box, so the zonal and global averages of amount do not exactly equal the product of zonal and global averages of f and awp.

 

Cloud Trends Over Land (1971-1996, maps shown within the land atlas)

      For each station, we examined the time series of individual yearly values for each season for each cloud type.  We computed a daytime trend for a season if we had a minimum of 75 observations in each of at least 15 years, spanning at least 20 years.  For night we required 50 observations.  A 5-degree grid box on land may contain several stations in different topographic regions, with correspondingly different cloud climatologies.  If one station began observations halfway through the period of record, an apparent but spurious trend would result if all observations in the grid-box for a given year were grouped together.  We therefore first perform a trend analysis for each individual station.  If a grid-box contains more than one station, we average their trends.  On the maps we display a trend if it exceeds its uncertainty, or if the uncertainty is less than 1% per decade.  The trends, as well as other interannual variations, are discussed by Warren et al. (2007).  Note that the computed trends reported here span the years 1971-96; trends computed over a different span of years will be different. 

 

Cloud Trends Over the Arctic

      Trends in cloud cover over the Arctic are computed for each 10o grid box for land, ocean, and combined land and ocean. Trends are for day/night average cloud amount. Detailed specifics of the averaging techniques used are given in Eastman and Warren (2010a and 2010b). Trends per 10o grid box represent a linear fit to a single time series of seasonal cloud cover anomalies. At least 50 observations must be present per season. Over land areas a 20-year span of data is required with at least 15 years of data present, while over the ocean a 30 year span is required with at least 25 years of data present in order to compute trends. The differing criteria are necessary because the span of the land data is 1971-2007, while the ocean data span 1954-2008. Combined land and ocean trends are only for the span 1971-2007. For an Arctic trend to be shown its magnitude must exceed its uncertainty or the uncertainty must be less than 2 %/decade.

 

Cloud Trends Over the Ocean

      Trends in cloud cover over the ocean are computed for the span 1954-2008. All trends represent the day/night average cloud amount. Eastman et. al. (2011) show spurious variations in the global time series of cloud cover for most types over the ocean. This variation has been removed in the form of a 10th degree polynomial fit. Trends of non-altered data are also available. In order for a trend to be computed for a particular grid box in a season, 20 individual years of data must be present spanning at least 30 years. 25 observations are required per season per grid box. Trends are shown only if their magnitude exceeds their uncertainty or if their uncertainty is less than 2 %/decade.


Cloud Trends Over Land (1971-2009, Land Trends link)

      Trends in cloud cover over land areas are computed for the span 1971-2009 (Eastman and Warren, 2013).  Values shown are for daytime only.  In order for a trend to be computed at a particular station in a season, 25 individual years of data must be present spanning at least 30 years. 75 observations are required per season per station. Trends are shown only if their magnitude exceeds their uncertainty or if their uncertainty is less than 2 %/decade.  Grid-box averages represent the average trend from all stations within that box.  Details on our averaging techniques are available in Warren et al. (2007).  The US, Canada, and New Zealand have stopped producing synoptic cloud cover reports due to the automation of cloud reports (ASOS).  Most stations in these countries lack sufficient spans of data to pass our criteria for plotting, so much of the land areas are now left blank in those countries.

 

Acknowledgments

      This work was supported by NSF Climate and Large-Scale Dynamics (Geosystems Database Infrastructure Program), and by NOAA Office of Global Programs (Climate Change Data and Detection Program), under grants ATM-99-08699, ATM-02-42124, ATM-06-30428, and AGS-10-21543 to the University of Washington, and ATM-99-08700, ATM-02-42128, and ATM-06-30396 to the University of Arizona.  The work was also supported by a computing grant from the National Center for Atmospheric Research.  Maps of cloud averages and this website were developed by Ignatius Rigor.  The trend maps were computed by Ryan Eastman.

 

 

References

 

Eastman, R, and S.G. Warren, 2010a:  Interannual variations of Arctic cloud types and their relation to sea ice.  Journal of Climate, 23, 4216-4232. Eastman and Warren 2010a 

 

Eastman, R, and S.G. Warren, 2010b:  Arctic cloud changes from surface and satellite observations.  Journal of Climate, 23, 4233-4242. Eastman and Warren 2010b 


Eastman, R, S.G. Warren, and C.J. Hahn, 2011:  Variations in cloud cover and cloud types over the ocean from surface observations, 1954-2008.  Journal of Climate,24, 5914-5934. Eastman et al. 2011


Eastman, R., S.G. Warren, 2013: A 39-Yr survey of cloud changes from land stations worldwide 1971–2009: Long-term trends, relation to aerosols, and expansion of the tropical belt. Journal of Climate, 26, 1286-1303. Eastman and Warren 2013


Eastman, R., S.G. Warren, 2014: Diurnal cycles of cumulus, cumulonimbus, stratus, stratocumulus, and fog from surface observations over land and ocean.. Journal of Climate, 27, 2386-2404. Eastman and Warren 2014

 

Hahn, C.J., and S.G. Warren, (1999, 2009):  Extended Edited Cloud Reports from Ships and Land Stations over the Globe, 1952-1996.  Numerical Data Package NDP-026C, Carbon Dioxide Information Analysis Center (CDIAC), Department of Energy, Oak Ridge, Tennessee (Documentation, 79 pp.).  http://cdiac.ornl.gov/epubs/ndp/ndp026c/ndp026c.html updated to 2008 (ocean) and 2009 (land)

 

Hahn, C.J., and S.G. Warren, 2003:  Cloud climatology for land stations worldwide, 1971-1996.  Numeric Data Package NDP-026D, Carbon Dioxide Information Analysis Center (CDIAC), Department of Energy, Oak Ridge, Tennessee, (Documentation, 35 pp.). http://cdiac.ornl.gov/epubs/ndp/ndp026d/ndp026d.html

 

Hahn, C.J.,  and S.G. Warren, 2007: A Gridded Climatology of Clouds over Land (1971-96) and Ocean (1954-97) Worldwide.  NDP-026E, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, Oak Ridge, TN.  (Documentation, 71 pp.).  http://cdiac.ornl.gov/epub/sndp/ndp026e/ndp026e.html

 

Hahn, C.J., S.G. Warren and J. London, 1994:  Cloud climatology for clouds over the globe from surface observations, 1982-1991: The total cloud edition.  Numeric Data Package NDP-026A, Carbon Dioxide Information Analysis Center (CDIAC), Department of Energy, Oak Ridge, Tennessee.

 

Hahn, C.J., S.G. Warren and J. London, 1995:  The effect of moonlight on observation of cloud cover at night, and application to cloud climatology.  J. Climate, 8, 1429-1446.

 

Warren, S.G., C.J. Hahn, J. London, R.M Chervin and R.L. Jenne, 1986:  Global Distribution of Total Cloud Cover and Cloud Type Amounts over Land.  NCAR Technical Note TN-273+STR, Boulder, CO, 29 pp. + 200 maps (also DOE/ER/60085-H1).  Available from Carbon Dioxide Information Analysis Center, Oak Ridge, Tennessee. Warren et al. 1986 

 

Warren, S.G., C.J. Hahn, J. London, R.M Chervin and R.L. Jenne, 1988:  Global Distribution of Total Cloud Cover and Cloud Type Amounts over the Ocean.  NCAR Technical Note TN-317+STR, Boulder, CO, 42 pp. + 170 maps (also DOE/ER-0406).  Available from Carbon Dioxide Information Analysis Center, Oak Ridge, Tennessee. Warren et al. 1988 

 

Warren, S.G., C.J. Hahn, and J. London, 1991.  Cloud hole over the United States?  Bull. Amer. Meteor. Soc., 72, 237-238.

 

Warren, S.G., R.M. Eastman, and C.J. Hahn, 2007:  A survey of changes in cloud cover and cloud types over land from surface observations, 1971-96.  J. Climate, 20, 717-738.

 

Woodruff, S.D., R.J. Slutz,  R.L. Jenne and P.M. Steurer, 1987:  A comprehensive ocean-atmosphere data set.  Bull. Amer. Meteor. Soc., 68, 1239-1250.

 

World Meteorological Organization, 1988:  Manual on Codes, Volume 1.  (WMO Publ. No. 306), WMO, Geneva.